# tokens Published articles for tokens. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Trajectory as the Teacher: Few-Step Discrete Flow Matching via Energy-Navigated Distillation DevFeed: [Trajectory as the Teacher: Few-Step Discrete Flow Matching via Energy-Navigated Distillation]() Original publisher: [Read original article]() Published: 2026-09-16T00:00:00Z Content type: article Language: en Sources: [Apple Machine Learning Research]() Topics: [text-generation](), [Algorithms](), [Language models](), [Inference]() Tags: [inference](), [language-models](), [performance](), [perplexity](), [research](), [text-generation](), [tokens]() ### AI overview The article introduces Trajectory-Shaped Discrete Flow Matching, a training method that guides intermediate trajectory decisions with an energy-based coherence measure. The authors argue that poor distillation trajectories, rather than insufficient student capacity, limit few-step generation. On a 170M-parameter language-modeling task, an 8-step student reportedly achieves lower perplexity than a 1,024-step teacher while reducing inference steps. ### Source excerpt Discrete flow matching generates text by iteratively transforming noise tokens into coherent language, but may require hundreds of forward passes. Distillation uses the multi-step trajectory to train a student to reproduce the process in a few steps. When the student underperforms, the usual explanation is insufficient capacity. We argue the opposite: the trajectory is the bottleneck, not the student. Each training trajectory is built through a chain of blind stochastic jumps with no evaluation of sequence quality; a single bad decision at an early midpoint propagates through subsequent steps... ## Dense vs. MoE Models: Active Parameters, Throughput, and When to Choose Each DevFeed: [Dense vs. MoE Models: Active Parameters, Throughput, and When to Choose Each]() Original publisher: [Read original article]() Author: Elizabeth Goodman Published: 2026-09-15T17:00:11Z Content type: article Language: en Sources: [NVIDIA Developer](), [NVIDIA Technical Blog]() Topics: [Mixture of Experts (MoE)](), [Nemotron](), [Routing (disambiguation)]() Tags: [agentic-ai-generative-ai](), [ai-inference](), [architecture](), [cost](), [data-center-cloud](), [deployment](), [llms](), [memory](), [mixture-of-experts-moe](), [models](), [moe](), [nemo](), [nemotron](), [performance](), [router](), [routing](), [tokens]() ### AI overview This article explains how dense and Mixture-of-Experts models activate parameters, compares their effects on throughput, memory cost, and serving complexity, and discusses when each architecture fits different deployment constraints. It uses Nemotron 3.5 Lightning as an example of an MoE model. ### Source excerpt How can a 30B-parameter model activate only 3B parameters per token, and still use the capacity of the larger model? Nemotron 3.5 Lightning illustrates the... ## Optimizing cost and latency with Amazon Bedrock prompt caching DevFeed: [Optimizing cost and latency with Amazon Bedrock prompt caching]() Original publisher: [Read original article]() Author: Daniel Abib Published: 2026-09-15T16:18:19Z Content type: tutorial Language: en Sources: [Artificial Intelligence]() Topics: [Amazon Bedrock](), [Caching](), [Latency](), [API](), [foundation-models](), [LangChain](), [Multi-tenancy]() Tags: [advanced-300](), [amazon-bedrock](), [api](), [caching](), [foundation-models](), [integration](), [langchain](), [latency](), [pricing](), [technical-how-to](), [tokens]() ### AI overview This AWS post explains how Amazon Bedrock prompt caching can reduce repeated input-token costs by up to 90 percent and lower time to first token when requests reuse the same context. It presents six scenarios using the Converse API, including document, system prompt, tool definition, mixed TTL, tenant-isolated, and LangChain caching. ### Source excerpt Prompt caching in Amazon Bedrock can cut input token costs by up to 90% when you repeatedly send the same context to foundation models. This post walks through six practical prompt caching scenarios using the Converse API: message content, system prompt, tool definition, mixed TTL, tenant isolation, and LangChain integration. ## Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine DevFeed: [Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine]() Original publisher: [Read original article]() Author: Tanya Lenz Published: 2026-09-14T16:39:15Z Content type: article Language: en Sources: [NVIDIA Developer](), [NVIDIA Technical Blog]() Topics: [Mixture of Experts (MoE)](), [Artificial Intelligence](), [GPU](), [Large Language Model](), [Nvidia](), [Routing (disambiguation)](), [GB200](), [Network](), [Python](), [deepseek](), [mixtral](), [qwen]() Tags: [agentic-ai-generative-ai](), [ai](), [communication](), [deepseek](), [developer-tools-techniques](), [gb200](), [gpu](), [mixtral](), [mixture-of-experts-moe](), [mlops](), [moe](), [networks](), [nvfp4](), [nvidia](), [performance](), [python](), [qwen](), [router](), [routing](), [tensors](), [token](), [tokens](), [training-ai-models]() ### AI overview This article explains how NVIDIA Transformer Engine and JAX kernel optimizations accelerate dropless Mixture of Experts (MoE) training. It describes bottlenecks from token routing, expert dispatch and gathering, all-to-all communication, and ragged expert matrix operations. In DeepSeek-V3 training on NVIDIA GB200, the optimized approach increased performance from 103 to 1,068 TFLOPS per GPU, a 10.4x improvement. ### Source excerpt Mixture of experts (MoE) has become one of the defining architectural trends in large-scale AI model training. DeepSeek, Qwen, and Mixtral are examples of MoE... ## Meta Open-Sources Astryx, Its Agent-Ready React Design System DevFeed: [Meta Open-Sources Astryx, Its Agent-Ready React Design System]() Original publisher: [Read original article]() Author: Bruno Couriol Published: 2026-09-13T23:39:00Z Content type: news Language: en Sources: [InfoQ]() Topics: [Meta](), [React](), [Open Source](), [CSS](), [Tooling](), [Accessibility](), [Model Context Protocol](), [User interface design](), [Web Development](), [Svelte](), [Flutter]() Tags: [accessibility](), [css](), [development](), [flutter](), [javascript](), [mcp](), [meta](), [meta-astryx-design-system](), [news](), [open-source](), [react](), [svelte](), [tokens](), [tooling](), [typescript](), [ui](), [web-development]() ### AI overview Meta has beta-released Astryx, an open-source React design system developed internally over eight years. It offers more than 150 accessible components, customizable CSS design tokens, precompiled StyleX-based styling, component code ejection, CLI tooling, and an MCP endpoint for developer and AI workflows. ### Source excerpt Meta recently announced the beta release of Astryx, an open-source React design system developed internally over eight years. Astryx builds on React 19 and StyleX to provide over 150 accessible UI components, customizable CSS design tokens, and dedicated CLI and MCP tooling -- for both engineers and AI agents. By Bruno Couriol ## Chip Huyen explains how to cut inference costs without new hardware DevFeed: [Chip Huyen explains how to cut inference costs without new hardware]() Original publisher: [Read original article]() Author: Tim Koopmans Published: 2026-09-13T15:00:00Z Content type: article Language: en Sources: [The New Stack]() Topics: [Inference Performance](), [Low-Latency Inference](), [model-deployment](), [Optimization](), [Frontier Model](), [AI Engineering](), [Latency](), [math]() Tags: [ai](), [ai-agents](), [ai-engineering](), [ai-infrastructure](), [frontier-model](), [inference](), [low-latency](), [metrics](), [optimization](), [performance](), [post-contributed](), [reasoning](), [scylladb](), [sponsor-scylladb](), [sponsored](), [sponsored-post-contributed](), [tokens]() ### AI overview Chip Huyen explains why inference costs can outweigh one-time frontier-model training costs and outlines ways to optimize inference without new hardware. The article emphasizes latency metrics such as time to first token, time per output token, end-to-end latency, and goodput, especially for reasoning models. ### Source excerpt Last October, the P99 conference -- the online gathering for developers focused on high-performance, low-latency applications -- featured a cracking The post Chip Huyen explains how to cut inference costs without new hardware appeared first on The New Stack. ## AP2 Protocol Explained: How Agent Payments Get Authorized DevFeed: [AP2 Protocol Explained: How Agent Payments Get Authorized]() Original publisher: [Read original article]() Author: Ayush Agarwal Published: 2026-09-13T00:00:00Z Content type: article Language: en Sources: [Dodo Payments Blog]() Topics: [Authorization](), [AI Agent](), [Artificial Intelligence](), [JSON Web Tokens](), [Google]() Tags: [agent](), [agentic](), [ai](), [authentication](), [cryptographic](), [google](), [payments](), [tokens]() ### AI overview A practical explanation of AP2, an authorization and evidence protocol for AI agent payment transactions. It describes cryptographic proof, verifiable digital credentials, SD-JWTs, and separate checkout and payment mandates. ### Source excerpt A practical guide to AP2, the Agent Payments Protocol: checkout and payment mandates, open versus closed authorization, verifiable digital credentials, and what it means for merchants. ## Fear Is Not an Argument DevFeed: [Fear Is Not an Argument]() Original publisher: [Read original article]() Author: Daniel Lemire Published: 2026-09-10T18:23:42Z Content type: opinion Language: en Sources: [Daniel Lemire]() Topics: [Artificial Intelligence](), [Large Language Model](), [Language models](), [ChatGPT](), [Tool](), [Embeddings]() Tags: [ai](), [chatgpt](), [embeddings](), [large-language-model](), [openai](), [statement](), [tokens](), [tools]() ### AI overview This opinion argues that fears about AI causing human extinction are vague and unfalsifiable, and compares them with earlier technological and social end-of-the-world predictions. It describes large language models as systems that process and generate tokens using fixed weights, while noting that connecting them to tools makes their capabilities more consequential. ### Source excerpt We are told that AI entities much like ChatGPT might soon kill us all. The statement is vague and unfalsifiable. It might be true, it might be false. People with credentials (e.g., Turing Award recipient Yoshua Bengio) believe it. Many still remember the Year-2000 bug. Our computers used two-digit coding for dates, and some software ... Continue reading Fear Is Not an Argument ## Firefox Is Getting Its Biggest Redesign in Years. Then Mozilla Got Cold Feet. DevFeed: [Firefox Is Getting Its Biggest Redesign in Years. Then Mozilla Got Cold Feet.]() Original publisher: [Read original article]() Author: Alex Harper Published: 2026-09-09T18:24:32Z Content type: opinion Language: en Sources: [Web Designer Depot]() Topics: [Firefox]() Tags: [browser](), [browser-design](), [browser-redesign](), [browser-ui](), [browser-ux](), [design](), [design-controversy](), [design-trends](), [features](), [firefox](), [firefox-features](), [firefox-interface](), [firefox-redesign](), [firefox-update](), [future-of-browsers](), [interface-design](), [mobile](), [mozilla](), [mozilla-firefox](), [open-source](), [product-design](), [themes](), [tokens](), [ui-design](), [ux](), [ux-design](), [visual-ui-design](), [web-browsers](), [web-design]() ### AI overview Firefox's Project Nova redesign is being softened after user feedback. Mozilla restored Compact Mode and reduced some of the boldest interface changes, while retaining refreshed components, icons, colors, themes, and rounded elements. The article argues that responsive iteration and user control are more valuable than following fashionable design trends. ### Source excerpt Firefox is getting its biggest makeover in years--but Mozilla has already backed away from some of its boldest ideas. Project Nova reveals what happens when modern design trends, millions of opinionated users, and the internet's unstoppable obsession with rounded corners collide. ## How Smart Model Routing Can Cut LLM Costs by 10X DevFeed: [How Smart Model Routing Can Cut LLM Costs by 10X]() Original publisher: [Read original article]() Author: ByteByteGo Published: 2026-09-09T15:30:26Z Content type: article Language: en Sources: [ByteByteGo]() Topics: [Model Routing](), [Large Language Model](), [API]() Tags: [cost](), [llm](), [model-routing](), [models](), [routing](), [routing-system](), [tokens]() ### AI overview This article explains how smart model routing can reduce LLM application costs by sending simple requests to less expensive models and difficult requests to more capable models. It notes that savings and response quality depend on request types, model price differences, and routing performance, and discusses router, cascading, semantic, and learned routing approaches. ### Source excerpt Cost reduction isn't a given. It also depends on the types of requests the application receives, the price difference between models, and how well the routing system performs. In this article, we are going to look at various aspects ## Optimize vLLM speculative decoding with FastMTP heads DevFeed: [Optimize vLLM speculative decoding with FastMTP heads]() Original publisher: [Read original article]() Author: Rahul Tuli Published: 2026-09-08T14:20:16Z Content type: article Language: en Sources: [Red Hat](), [Red Hat Developer]() Topics: [vllm](), [Inference](), [Large Language Model](), [Fine-tuning](), [model-deployment](), [deepseek](), [qwen]() Tags: [ai-inference](), [artificial-intelligence](), [data](), [deepseek](), [developer-tools](), [fine-tuning](), [hardware](), [inference](), [large-language-model](), [production](), [qwen](), [tokens](), [vllm]() ### AI overview This article explains how FastMTP-style fine-tuning improves vLLM speculative decoding. It describes using native multi-token prediction heads as speculators, adapting a single head for recursive multi-step drafting, extracting weights from verifier checkpoints, and producing vLLM-ready checkpoints without training from scratch. ### Source excerpt Autoregressive decoding makes large language model (LLM) inference memory-bandwidth bound: every token needs 1 full forward pass over billions of parameters, so the hardware spends most of its time moving weights rather than computing. MTP is a training objective: models like the DeepSeek and Qwen families learn to predict several future tokens at each position, which improves their data efficiency and quality. The post Optimize vLLM speculative decoding with FastMTP heads appeared first on Red Hat Developer. ## How engineering teams can measure AI coding ROI with cost attribution and outcome metrics DevFeed: [How engineering teams can measure AI coding ROI with cost attribution and outcome metrics]() Original publisher: [Read original article]() Author: Kelsey Rosen Published: 2026-09-08T00:00:00Z Content type: article Language: en Sources: [Harness Blog]() Topics: [ai-coding](), [Ground truth / benchmark quality](), [coding](), [Finance]() Tags: [ai](), [ai-coding](), [analysis](), [blog](), [coding](), [cost](), [datadog](), [engineering](), [governance](), [harness](), [incident](), [lifecycle](), [measurement](), [metrics](), [numbers](), [production](), [team](), [tokens](), [tooling](), [workflow]() ### AI overview A Harness panel discussion outlines ways engineering organizations can measure AI coding ROI. The panel recommends attributing costs to specific work, combining throughput metrics with qualitative evidence, and using governance and process changes to interpret results. ### Source excerpt Harness panelists share practical strategies for measuring AI coding ROI, from granular cost attribution and lifecycle metrics to budgeting, governance. | Blog ## How to Implement OIDC Back-Channel Logout DevFeed: [How to Implement OIDC Back-Channel Logout]() Original publisher: [Read original article]() Author: WorkOS Published: 2026-09-08T00:00:00Z Content type: tutorial Language: en Sources: [WorkOS Blog]() Topics: [OpenID connect (OIDC)](), [Single sign-on (SSO)](), [JSON Web Tokens](), [App](), [Server]() Tags: [how-to](), [implement](), [jwt](), [oidc](), [server](), [sso](), [token](), [tokens]() ### AI overview This tutorial explains OIDC back-channel logout, including provider discovery, endpoint registration, form-encoded logout requests, and logout-token validation. It also describes how session-specific and user-wide logout work and notes limitations acknowledged by the specification. ### Source excerpt It is the one standard that can end a session your app already issued, it has been final since 2022, and almost nobody implements it. Here is the whole mechanism, the validation your endpoint owes, and the two limitations the spec admits to itself. ## How AI Tools Accelerate the Reward Loop of Building DevFeed: [How AI Tools Accelerate the Reward Loop of Building]() Original publisher: [Read original article]() Author: Jon C. Phillips Published: 2026-09-07T02:54:00Z Content type: opinion Language: en Sources: [Jon C. Phillips]() Topics: [Artificial Intelligence]() Tags: [ai](), [articles](), [audience-building](), [code](), [design](), [digital-products](), [music](), [no-natural-limit](), [photography](), [product-engineering](), [side-projects](), [tokens](), [web-development]() ### AI overview The author argues that building has long been driven by an unpredictable reward loop, but AI tools have shortened the time from idea to working product, making that loop run faster and potentially changing the nature of the hobby. ### Source excerpt A friend and I were comparing. They'd done about 10 billion tokens in the same period and neither of us found that alarming, really. Which is probably the first thing worth noticing. And we probably all know people running way higher. ## How to build an AI-powered app on Webflow Cloud with OpenAI and Supabase DevFeed: [How to build an AI-powered app on Webflow Cloud with OpenAI and Supabase]() Original publisher: [Read original article]() Author: Ismail Ajagbe Published: 2026-09-04T00:00:00Z Content type: tutorial Language: en Sources: [Webflow Blog]() Topics: [AI Chat](), [Artificial Intelligence](), [Next.js](), [Workers](), [API](), [Node.js](), [Command-line interface](), [Deployment]() Tags: [ai](), [api](), [article](), [cli](), [cloud](), [cloudflare-workers](), [deployment](), [guides](), [how-to](), [next-js](), [node-js](), [openai](), [serverless](), [storage](), [tokens](), [webflow]() ### AI overview Learn how to build a streaming AI chat app on Webflow Cloud with OpenAI and Supabase, including Next.js route handlers, Cloudflare Workers setup, browser token streaming, and optional conversation history storage. ### Source excerpt Learn how to build a streaming AI chat app on Webflow Cloud with OpenAI and Supabase. ## Attention Mechanisms in LLMs, clearly explained DevFeed: [Attention Mechanisms in LLMs, clearly explained]() Original publisher: [Read original article]() Author: Avi Chawla Published: 2026-09-03T20:05:32Z Content type: tutorial Language: en Sources: [Daily Dose of Data Science]() Topics: [Large Language Model](), [Cache]() Tags: [cache](), [llms](), [tokens]() ### AI overview A tutorial explaining why attention is used in language models, how long-range dependencies are handled, and why storing attention state for long sequences and large batches can exhaust GPU memory. It introduces attention variants including Multi-Query Attention, Grouped-Query Attention, and Multi-Head Latent Attention. ### Source excerpt Everything you need to understand how attention works, why the KV cache is the bottleneck, and what every attention variant is actually solving. ## Five Reasons Why Your Auth0 Permissions Are Not in your Access Token DevFeed: [Five Reasons Why Your Auth0 Permissions Are Not in your Access Token]() Original publisher: [Read original article]() Author: Carla Urrea Stabile Published: 2026-09-03T00:00:00Z Content type: tutorial Language: en Sources: [Auth0 Blog]() Topics: [Auth0](), [Authorization](), [JSON Web Tokens](), [AWS Identity and Access Management (IAM)]() Tags: [auth0](), [authorization](), [how-to](), [jwt](), [tokens]() ### AI overview A tutorial explaining five common reasons Auth0 RBAC permissions may be missing from an access token, including an omitted audience parameter, disabled permission settings, and incomplete role assignments. ### Source excerpt Decoding your Auth0 JWT and seeing an empty permissions array? Here are 5 common reasons your Auth0 RBAC permissions are missing and how to fix them fast. ## How Coinbase used Code Connect to guide agents and shrink token costs DevFeed: [How Coinbase used Code Connect to guide agents and shrink token costs]() Original publisher: [Read original article]() Author: Jenny Xie Published: 2026-09-02T20:28:00Z Content type: article Language: en Sources: [Figma Blog]() Topics: [Figma](), [Model Context Protocol](), [AI-assisted coding](), [Code quality](), [Accessibility](), [Cryptocurrency]() Tags: [accessibility](), [agents](), [code](), [coding](), [coinbase](), [cost](), [cryptocurrency](), [figma](), [mcp](), [mcp-server](), [tokens](), [web]() ### AI overview The Coinbase Design System team evaluated Code Connect in Figma's MCP server with coding agents. The article reports improved adherence to the design system, better code quality, and an average 22.5% reduction in token costs when converting Figma designs into code. ### Source excerpt The Coinbase Design System team put Code Connect to the test against coding agents--and found that it boosted design system adherence while cutting token costs by an average of 22.5%. ## GLM-5.3 is 50% off through DigitalOcean on AI Gateway DevFeed: [GLM-5.3 is 50% off through DigitalOcean on AI Gateway]() Original publisher: [Read original article]() Author: Jerilyn Zheng Published: 2026-09-02T00:00:00Z Content type: release Language: en Sources: [Vercel News]() Topics: [Vercel](), [Digital Ocean](), [Language models](), [long-context](), [AI-assisted coding](), [Claude Code](), [codex](), [cursor]() Tags: [ai-gateway](), [claude-code](), [codex](), [coding-agents](), [context-window](), [cursor](), [digitalocean](), [language-models](), [model](), [tokens](), [vercel]() ### AI overview Vercel announces a 50% discount on GLM-5.3 through DigitalOcean on AI Gateway until September 8. The article explains the temporary promo model name, standard provider routing, model limits, spend tracking, and setup for coding agents. ### Source excerpt GLM-5.3 is 50% off on AI Gateway through Tuesday, September 8, in partnership with DigitalOcean. How to use the model during the offer period Using the promo name (zai/glm-5.3-promo-50) gets the discounted rate. It routes only to DigitalOcean, with no fallback to another provider, and it stops serving when the offer ends. Using the standard name (i.e., zai/glm-5.3) with provider options to sort DigitalOcean as the preferred provider keeps working after September 8 and routes across every provider that serves the model, at their usual rates. Because the promo name goes away when the offer ends, treat it as something you switch on for the window rather than hardcode. To keep the standard name in your code instead, pin the provider with order: ['digitalocean'] under providerOptions.gateway, which prefers DigitalOcean and falls back to the others if it cannot serve the request. GLM-5.3 takes text input, with a 1M token context window and a maximum output of 128K tokens. Discounted requests appear in your spend dashboard and carry a trace like any other request. Try GLM-5.3 in the model playground. To use it in a coding agent, see the coding agents guide, then run vercel ai-gateway coding-agents setup to connect agents like Claude Code, Codex, OpenCode, Cursor, Pi, and more and select zai/glm-5.3-promo-50 inside the agent. You can view all language models available on AI Gateway. Read more ## Refresh token behavior across fourteen providers DevFeed: [Refresh token behavior across fourteen providers]() Original publisher: [Read original article]() Author: WorkOS Published: 2026-08-31T00:00:00Z Content type: article Language: en Sources: [WorkOS Blog]() Topics: [OAuth](), [API]() Tags: [providers](), [retry](), [rotation](), [token](), [tokens]() ### AI overview This article compares documented OAuth refresh-token behavior across fourteen providers, including token rotation, expiration reporting, grace periods, and revocation on reuse. It explains how rotation affects persistence and retry handling, with examples including Google, Box, Atlassian, Dropbox, Asana, GitHub, and Salesforce. ### Source excerpt Which providers rotate refresh tokens, which return expires_in, which give you a grace period, and which revoke on reuse. One row per provider, verified against provider documentation in August 2026. ## Type-Safe Authentication in Rust with Auth0 and Axum DevFeed: [Type-Safe Authentication in Rust with Auth0 and Axum]() Original publisher: [Read original article]() Author: Aniket Bhattacharyea Published: 2026-08-31T00:00:00Z Content type: tutorial Language: en Sources: [Auth0 Blog]() Topics: [Authentication](), [Auth0](), [Rust](), [Command-line interface](), [OAuth 2.0](), [Authorization](), [API]() Tags: [api](), [auth0](), [authentication](), [authorization](), [cli](), [oauth](), [oauth-2-0](), [rust](), [tokens](), [tutorial]() ### AI overview A tutorial showing how to build a Rust CLI that uses Auth0's OAuth 2.0 Device Authorization Flow, then create an Axum API that validates the resulting access token before allowing access to a protected route. ### Source excerpt Secure your Rust CLI applications with Auth0 Device Code flow and protect your Axum backend endpoints by validating JWTs with Rust's type safety. ## Kubernetes v1.37: Pod Certificates and Cluster Trust Bundles DevFeed: [Kubernetes v1.37: Pod Certificates and Cluster Trust Bundles]() Original publisher: [Read original article]() Author: Taahir Ahmed Published: 2026-08-28T18:30:00Z Content type: article Language: en Sources: [Kubernetes Blog]() Topics: [Kubernetes](), [Authentication](), [JSON Web Tokens](), [TLS (Transport Layer Security)](), [Cloud]() Tags: [authentication](), [blog](), [cloud](), [cryptographic](), [json](), [kubernetes](), [post](), [tls](), [tokens]() ### AI overview Kubernetes 1.37 introduces generally available Pod Certificates and Cluster Trust Bundles, adding built-in X.509 certificate issuance for TLS and mTLS. The article explains how this production identity mechanism relates to existing service account JWTs and addresses the security limitations of bearer tokens. ### Source excerpt Pod Certificate / Cluster Trust Bundles Blog Post Kubernetes brings a wealth of features that make it easy to run your production workloads securely and reliably. While aspects like scheduling, health checks and resource limits are probably at the front of your mind, one other important feature of Kubernetes is production identity -- how your workload can authenticate to other systems in order to do its job. Up until now, the primary production identity mechanism built into Kubernetes has been service account JWTs (JSON Web Tokens). These are cryptographically-signed tokens, issued by the control plane of your cluster, that let anyone in the world understand who is calling when your workload uses them. In Kubernetes 1.37, the foundations of a new built-in production identity technology have gone GA. Pod Certificates (and the closely-associated Cluster Trust Bundles) build X.509 certificate issuance for TLS and mTLS directly into core Kubernetes. Why? Service account JWTs have a lot going for them: They are built directly into Kubelet, and work pretty magically. They are written to your workload container's filesystem before your workload starts up, and automatically kept up to date. The issuance system follows least-privilege principles; the node restriction admission plugin ensures that tokens can only be requested by the Kubelet that is actually currently running your pod. They can be federated, allowing you to use them to authenticate to other systems outside of Kubernetes. Service account JWTs underpin the pod-to-cloud authentication store for all of the largest cloud providers, and have widespread support across many additional services and software packages. If it can understand JWTs, you can authenticate to it with a service account token. However, service account JWTs have one big downside -- they are bearer tokens. With bearer tokens, if you have the token, then you are the identity asserted by the token. And since you necessarily have to hand copies of the J ## Three ways to let an AI agent call third-party APIs on behalf of a user DevFeed: [Three ways to let an AI agent call third-party APIs on behalf of a user]() Original publisher: [Read original article]() Author: WorkOS Published: 2026-08-27T00:00:00Z Content type: tutorial Language: en Sources: [WorkOS Blog]() Topics: [AI Agent](), [OAuth](), [Security]() Tags: [agent](), [ai-agent](), [apis](), [hubspot](), [linear](), [oauth](), [patterns](), [proxy](), [security](), [slack](), [third-party](), [tokens]() ### AI overview This article explains three patterns for letting an AI agent access third-party APIs on a user's behalf: storing OAuth credentials yourself, fetching tokens at runtime, or using a proxy that keeps tokens out of the agent-controlled runtime. It compares their operational and security implications and recommends choosing based on where the code runs and the system's requirements. ### Source excerpt Store the token yourself, fetch it at runtime, or never hold it at all. Where the credential ends up in each pattern, what each one costs, and how to pick without guessing. ## How Speculative Decoding Can Make LLM Generation 2-3 Times Faster DevFeed: [How Speculative Decoding Can Make LLM Generation 2-3 Times Faster]() Original publisher: [Read original article]() Author: ByteByteGo Published: 2026-08-26T15:30:34Z Content type: tutorial Language: en Sources: [ByteByteGo]() Topics: [Large Language Model](), [text-generation](), [GPU]() Tags: [article](), [data-center](), [decoding](), [generation](), [gpu](), [how-to](), [llms](), [tokens]() ### AI overview This tutorial explains speculative decoding, in which a smaller model proposes candidate tokens and a larger model evaluates them in a single forward pass. It covers autoregressive generation, GPU utilization, candidate acceptance and rejection, output-quality preservation, acceptance rates, draft sources, and when the technique may stop helping. ### Source excerpt In this article, we will look at how speculative decoding works. [Next page]()